# textCNN_XLNET_Cross Validation
## Cross validation 前
* 原本使用全部資料 => training
* 將所有非正常與正常 1:1 => testing

#### 問題
testing 的資料都在 training data 出現過，可能造成 overting

## 使用 Cross validation
fold = 5
data: word_labeled_zh_200918_02:17CC.csv
時間: 6.5 hr
**資料描述**

#### CV 結果 : 命中率雖高，但效果不好
-------------------------------------------
-------------------------------------------
-------------------------------------------
-------------------------------------------

### 推測原因
資料不平衡且不平均分布

# Update 2020/12/27
### data information
* training data: word_label_data_remove_long_word_201223.csv
* fold: 5 (Train:Test = 8:2)
### 效果
#### Fold 0 (**Accuracy: 0.936**)
* model: **text_cnn_best_0.9362186788154897_LR0.001_BATCH100_EPOCH100**



* predict error (err: 465): https://docs.google.com/spreadsheets/d/1JNuHyBOSBIFLDATdaWymgu30GHRJvtpigSLGUlC_GSg/edit?usp=sharing

---

#### Fold 1 (**Accuracy: 0.933**)
* model: **text_cnn_best_0.9335387913707625_LR0.001_BATCH100_EPOCH100**



* predict error (err:473): https://docs.google.com/spreadsheets/d/1dxq_1wdXgpiiCCqsWOrFsZtX5Le391bMMNd7xl0aYg8/edit?usp=sharing

---

#### Fold 2 (**Accuracy: 0.928**)
* model: **text_cnn_best_0.9287149939702533_LR0.001_BATCH100_EPOCH100**



* predict error (err:536): https://docs.google.com/spreadsheets/d/1AH8HBkOyNcMRAXzGeBv-ZUt6Nv3xx6Pc73Cs4qE9QRk/edit?usp=sharing

---

#### Fold 3 (**Accuracy: 0.937**)
* model: **text_cnn_best_0.9371482176360225_LR0.001_BATCH100_EPOCH100**



* predict error (err: 451): https://docs.google.com/spreadsheets/d/13P8Waolh6jKTpKIPDpV701RZsEz1Cv_mlOBzCHqi_60/edit?usp=sharing

---
#### Fold 4 (**Accuracy: 0.926**)
* model: **text_cnn_best_0.9261592066470116_LR0.001_BATCH100_EPOCH100**



* predict error (err: 537): https://docs.google.com/spreadsheets/d/1Kr8XNZwokP58-EVb41BQy5tv1kzF9FNWXs9nJeS8iVk/edit?usp=sharing

# Update 2021/1/3
## 調整 weight
**效果僅微幅上升，混淆矩陣效果仍然不佳**

#### fold 0
* model: text_cnn_best_0.9385647216633132_LR0.001_BATCH100_EPOCH100

#### fold 1

* model: text_cnn_best_0.9362843729040912_LR0.001_BATCH100_EPOCH100

#### fold 2

* model: text_cnn_best_0.9299704856452912_LR0.001_BATCH100_EPOCH100

#### fold 3

* model: text_cnn_best_0.9331902334317145_LR0.001_BATCH100_EPOCH100

#### fold 4

* model: text_cnn_best_0.9369466058492085_LR0.001_BATCH100_EPOCH100

### code
https://colab.research.google.com/drive/1tysuKunII1paJo6VXZtDcW6Qq5bxacJz?usp=sharing
###### tags: `Progress Report`